Predictive Therapy Neural Stimulation System

Abstract

Systems, devices, and methods for electrically stimulating a peripheral nerve(s) to treat various conditions, as well as signal processing systems and methods for enhancing diagnostic and treatment protocols associated therewith, are disclosed.

Description

[0001] Citation of Related Applications

[0002] This application claims the benefit of U.S. Provisional Application No. 62 / 736,968, filed Sep. 26, 2018, as a non - provisional application under 35 U.S.C § 119(e), which is hereby incorporated by reference in its entirety. Technical Field

[0003] Embodiments of the present invention generally relate to systems, devices, and methods for stimulating nerves, and more particularly to systems, devices, and methods for electrically stimulating peripheral nerve(s) to treat various conditions, as well as signal processing systems and methods for enhancing diagnostic and treatment protocols related thereto. Background Art

[0004] Peripheral nerves can be neuromodulated using a variety of treatment modalities. For example, electrical energy can be transcutaneously delivered via electrodes on the skin surface by a nerve stimulation system to stimulate peripheral nerves (such as the median nerve, radial nerve, and / or ulnar nerve in the upper limb; the tibial nerve, saphenous nerve, and / or peroneal nerve in the lower limb; or the auricular vagus nerve, tragus, trigeminal nerve, or cranial nerves on the head or ear, as non - limiting examples). Stimulating these nerves has been shown to provide therapeutic benefits across a variety of diseases, including but not limited to movement disorders (including but not limited to essential tremor, Parkinson's tremor, orthostatic tremor, and multiple sclerosis), urological conditions, gastrointestinal conditions, heart diseases, and inflammatory diseases, mood disorders (including but not limited to depression, bipolar disorder, dysthymia, and anxiety disorders), pain syndromes (including but not limited to migraine and other headaches, trigeminal neuralgia, fibromyalgia, complex regional pain syndrome), and so on. A variety of symptoms (such as tremors) can be treated by some form of percutaneous, trans - cutaneous, or other implanted form of peripheral nerve stimulation. There is a need for a wearable system with a compact, ergonomic form factor to enhance the efficacy, compliance, and comfort of using the device. Summary of the Invention

[0005] In some embodiments, a neuromodulation device is disclosed herein according to any one or more of the embodiments described in the present disclosure.

[0006] A system and / or method for predicting a clinical score is also disclosed herein according to any one or more of the embodiments described in the present disclosure.

[0007] A system and / or method for predicting a response or lack thereof to a therapy is further disclosed herein according to any one or more of the embodiments described in the present disclosure.

[0008] In some embodiments, disclosed herein are wearable nerve stimulation devices for transcutaneously stimulating one or more peripheral nerves of a user. The device may include one or more electrodes configured to generate an electrical stimulation signal. The device may further include one or more sensors configured to detect motion signals, wherein the one or more sensors are operably connected to the wearable nerve stimulation device; the device may also include one or more hardware processors. The one or more hardware processors may receive the raw signal in the time domain from the one or more sensors. The one or more hardware processors may separate the raw signal into a plurality of frames. In some cases, for each of the plurality of frames, the one or more hardware processors may transform the raw signal into the frequency domain. Further, the one or more hardware processors may calculate a first energy in a first frequency band of the transformed signal for each respective frame. The one or more hardware processors may also calculate a second energy in a second frequency band of the transformed signal for each respective frame, wherein the second frequency band includes a first frequency corresponding to tremor. The one or more hardware processors may determine motion artifacts in each frame based on a comparison of the first energy and the second energy. The one or more hardware processors may combine the frames based on the determination of motion artifacts for each in the frames. In addition, the one or more hardware processors may extract features from the combined frames in the time domain or the frequency domain. The one or more hardware processors may determine rules based on the extracted features. In some cases, the one or more hardware processors may determine the result of a nerve stimulation therapy based on the application of the determined rules on operational data.

[0009] In some cases, the sensors are operably attached to the wearable nerve stimulation device.

[0010] In some cases, the raw signal is related to the tremor activity of the user.

[0011] In some cases, the wearable nerve stimulation device may include one or more end effectors that may generate a stimulation signal other than the electrical stimulation signal.

[0012] In some cases, the stimulation signal other than the electrical stimulation signal is a vibration stimulation signal.

[0013] In some cases, the sensor includes an IMU. The IMU may include one or more of a gyroscope, an accelerometer, and a magnetometer.

[0014] In some cases, the second frequency band is between about 4 Hz and about 12 Hz.

[0015] In some cases, the second frequency band is between about 3 Hz and about 8 Hz.

[0016] In some cases, the features can include any one or more of the following: amplitude, bandwidth, area under the curve (e.g., power), energy in frequency bins, peak frequency, or ratio between frequency bands.

[0017] In some cases, the features include one or more of the kinematic features. The kinematic features can include the regularity, amplitude, and shape of the signal.

[0018] In some cases, the features can include any one or more of the following: amplitude or power spectral density (“PSD”) at the peak tremor frequency, total amplitude or PSD in a frequency band approximately 2.75 Hz wide around the peak tremor frequency, total amplitude or PSD in a frequency band between about 4 and about 12 Hz, or total amplitude or PSD in a frequency band around the peak tremor frequency selected only from the pre-stimulus spectrum.

[0019] In some cases, the features include time-domain features. In some cases, the features include frequency-domain features. In some cases, the features include a combination of time-domain and frequency-domain features.

[0020] In some cases, the features include any one or more of the following: approximate entropy, displacement, curve fitting, functional PCA, filtering, mean, median, or at least one in the time domain range.

[0021] In some embodiments, disclosed herein is a wearable device for transcutaneously modulating one or more peripheral nerves of a user. The device can include one or more end effectors configured to generate a stimulation signal. The device can further include one or more sensors configured to detect motion signals, wherein the one or more sensors are operably connected to the wearable device. The device can include one or more hardware processors. The one or more hardware processors can receive the raw signal in the time domain from the one or more sensors. The one or more hardware processors can separate the raw signal into a plurality of frames. In some cases, for each of the plurality of frames, the one or more hardware processors can perform one or more of the following operations: transform the raw signal into the frequency domain; calculate a first energy in a first frequency band of the transformed signal for each frame; calculate a second energy in a second frequency band of the transformed signal for each frame, wherein the second frequency band includes a first frequency corresponding to tremor; determine motion artifacts in each frame based on a comparison of the first energy and the second energy; combine the frames based on the determination of motion artifacts for each in the frames; extract features from the combined frames in the time domain or frequency domain; determine a rule based on the extracted features; and determine one or more of a clinical score and a nerve stimulation therapy outcome based on an application of the determined rule to the operational data.

[0022] In some cases, the one or more hardware processors may determine a first calibration frequency for a first stimulation therapy during a first activity, and a second different calibration frequency for a second stimulation therapy during a second different activity. In some cases, the first calibration frequency is within about 3 Hz of the second calibration frequency.

[0023] In some embodiments, methods are disclosed for transcutaneously stimulating one or more peripheral nerves of a user. The method may include generating an electrical stimulation signal via a pulse generator to one or more electrodes positioned on the skin surface of the user. The method may further include detecting a motion signal via one or more sensors. The method may include processing the motion signal via a hardware processor. Processing of the motion signal may include any one or more of the following operations: receiving the raw signal in the time domain from the one or more sensors; separating the raw signal into a plurality of frames. In some cases, for some or all of the plurality of frames, the method includes one or more of the following operations: transforming the raw signal into the frequency domain; calculating a first energy in a first frequency band of the transformed signal for each frame; calculating a second energy in a second frequency band of the transformed signal for each frame, where the second frequency band includes a first frequency corresponding to tremor; determining motion artifacts in each frame based on a comparison of the first energy and the second energy; combining the frames based on the determination of motion artifacts for each in the frames; extracting features from the combined frames in the time domain or the frequency domain; determining rules based on the extracted features; and determining the result of a nerve stimulation therapy based on the application of the determined rules to the operational data.

[0024] In some cases, the method may include identifying a user with essential tremor.

[0025] In some cases, the method may include identifying a user with Parkinson's disease.

[0026] In some cases, the method may include: determining a first calibration frequency for a first stimulation therapy during a first activity, and a second different calibration frequency for a second stimulation therapy during a second different activity.

[0027] In some cases, the first calibration frequency is within about 3 Hz of the second calibration frequency.

[0028] In some cases, the first activity is selected from: movement, drawing, posture holding, and tipping.

[0029] In some embodiments, methods for treating a patient having a tremor using percutaneously applied peripheral nerve stimulation therapy are disclosed herein. The method can include instructing the patient to perform a first tremor-inducing activity to elicit a first induced tremor. The method can include measuring movement of the patient's limb using a wearable biomechanical sensor to characterize the frequency of the first induced tremor. The method can include electrically stimulating an afferent peripheral nerve using a first set of stimulation parameters, at least in part based on the frequency of the first induced tremor. In some instances, after electrically stimulating the afferent peripheral nerve, the patient is instructed to perform a second tremor-inducing activity different from the first tremor-inducing activity to elicit a second induced tremor. The method can further include measuring movement of the patient's limb using the wearable biomechanical sensor to characterize the frequency of the second induced tremor. The method can also include electrically stimulating the afferent peripheral nerve using a second set of stimulation parameters, at least in part based on the frequency of the second induced tremor.

[0030] In some embodiments, methods for calibrating a nerve stimulation device are disclosed herein. The method can include collecting motion data at a sampling rate over a first time period, where the motion data includes motion corresponding to a plurality of axes. The method can include separating the collected motion data into a plurality of windows. The method can further include performing a frequency transform on each of the plurality of windows for each of the plurality of axes. The method can also include combining the frequency transform spectra of the plurality of axes for each of the plurality of windows. The method can further include combining the respective spectra from each of the plurality of windows into a calibration spectrum. The method can also include determining a peak from the calibration spectrum. The method can include performing calibration based on the determined peak.

[0031] In some instances, the method can include averaging the plurality of windows to generate a calibration spectrum.

[0032] In some instances, the method can include discarding one or more of the plurality of windows based on a detection of an artifact or noise.

[0033] In some instances, the first time period is about 12 seconds.

[0034] In some instances, the length of each of the plurality of windows is 2.4 seconds.

[0035] In some embodiments, methods for predicting the therapeutic efficacy of neural stimulation on a user are disclosed herein. The method may include determining a first feature, the first feature including a first frequency in the 4 - 12 Hz band having the highest power. The method may further include determining a second feature, the second feature including a first power at a peak of the frequency having the highest power in the 4 - 12 Hz band. The method may further include determining a third feature, the third feature including an average power in a positive or negative 1.5 Hz window centered on the frequency having the highest power in the 4 - 12 Hz band. The method may include determining a fourth feature, the fourth feature including a sum of powers in a positive or negative 1.5 Hz window centered on the frequency having the highest power in the 4 - 12 Hz band. The method may include determining a fifth feature, the fifth feature including a total power in the 4 - 12 Hz band. The method may include determining a sixth feature, the sixth feature including an entropy of the power spectral density in the 4 - 12 Hz band. The method may further include determining a seventh feature including a Q - factor, the Q - factor being the peak frequency divided by the frequency range where the spectral power is higher than 50% of the peak power. The method may further include determining an eighth feature including the temporal regularity of time - series data. In some cases, the method includes determining only some of the above eight features. The method may further include predicting the therapeutic efficacy based on the application of respective weights corresponding to any one or more of the first, second, third, fourth, fifth, sixth, seventh, and eighth features.

[0036] In some cases, the respective weights are calculated based on training using a machine - learning model. In some cases, the therapeutic efficacy includes a clinical rating. In some cases, the therapeutic efficacy includes a probability. In some cases, the therapeutic efficacy includes the time until the next treatment is required.

[0037] In some embodiments, methods for predicting the therapeutic efficacy of neural stimulation on a user are disclosed herein. The method may include collecting motion data at a sampling rate over a first time period, where the motion data includes motion corresponding to multiple axes. The method may further include separating the collected motion data into a plurality of windows. The method may further include performing a frequency transform on each of the plurality of windows for each of the multiple axes. The method may further include determining a peak frequency for each of the plurality of windows for each of the multiple axes. The method may further include: calculating a measure of variability of each peak - frequency window among the plurality of windows, the variability of the frequency - transform spectrum for each of the plurality of windows for each of the multiple axes. The method may further include predicting the therapeutic efficacy based on the magnitude of the measure of variability. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1A A block diagram illustrating an example neuromodulation (e.g., neural stimulation) device is shown.

[0039] Figure 1B A block diagram illustrating an embodiment of a controller, which can be implemented with the hardware components described with respect to Figure 1A Description.

[0040] Figure 1C Schematically illustrates embodiments of a neuromodulation device and a base station.

[0041] Figure 2 A block diagram illustrating an embodiment of a controller, which can be implemented with the hardware components described with respect to Figure 1A or Figure 1B Description.

[0042] Figure 3 A flowchart illustrating an embodiment of a process for calibrating a nerve stimulation device.

[0043] Figure 3A A flowchart illustrating a non - limiting embodiment of a process for performing window calibration.

[0044] Figure 3B Is a diagram showing that in some cases, the tremor frequency of a patient can vary across different tasks.

[0045] Figure 4 And Figure 4A A flowchart illustrating an embodiment of a process for collecting data from an IMU and processing the collected data.

[0046] Figure 5A A flowchart illustrating an embodiment of a process for generating rules for determining the outcome of a nerve stimulation therapy.

[0047] Figure 5B Illustrates an example combined spectrum including example - extracted features.

[0048] Figure 6 A schematically illustrates a wearable neuromodulation device.

[0049] Figure 6 B illustrates an embodiment of a stimulation waveform.

[0050] Figure 6 C illustrates a stimulation assessment at different time points.

[0051] Figure 6 D illustrates features extracted from a power spectral density chart.

[0052] Figure 7 A - Figure 7 D illustrate a reduction in tremor kinematics of sensor measurements regarding a nerve modulation therapy.

[0053] Figure 8 A- Figure 8 D illustrates the correlation graph between the average peak tremor power and the clinical visual assessment.

[0054] Figure 9 A- Figure 9 D illustrates that the kinematics measured by the sensor can predict the clinical assessment across tasks.

[0055] Figure 10 It is a graph illustrating the correlation between motion data (e.g., accelerometer) and the clinical scale.

[0056] Figure 11 It demonstrates that data collected over a longer time period can improve the prediction of patient response.

[0057] Figure 12 It further demonstrates how the patient response on the first day / week / month of therapy can predict the response across all segments based on the average results during the study period.

[0058] Figure 13 It illustrates that during neuromodulation therapy, the efficacy of neuromodulation therapy is high for patients both on and off medication (e.g., tremor medication).

[0059] Figure 14 It is a graph illustrating that the measured therapy outcome of the patient is correlated with the improvement in kinematics measured by the sensor.

[0060] Figure 15 A- Figure 15 D illustrates an example of an endpoint for sensor metrics according to some embodiments. Detailed embodiments

[0061] This disclosure relates to devices configured to provide neuromodulation (e.g., neural stimulation). The neuromodulation (e.g., neural stimulation) devices provided herein can be configured to stimulate a user's peripheral nerves. The neuromodulation (e.g., neural stimulation) devices can be configured to transcutaneously transmit one or more neuromodulation (e.g., neural stimulation) signals across the user's skin. In many embodiments, the neuromodulation (e.g., neural stimulation) device is a wearable device configured to be worn by the user. The user can be a human, another mammalian, or other animal user. The neuromodulation (e.g., neural stimulation) system can also include signal processing systems and methods for enhancing associated diagnostic and treatment protocols. In some embodiments, the neuromodulation (e.g., neural stimulation) device is configured to be wearable on the user's upper limb (e.g., the user's wrist, forearm, arm, and / or finger(s)). In some embodiments, the device is configured to be wearable on the user's lower limb (e.g., ankle, calf, knee, thigh, foot, and / or toe). In some embodiments, the device is configured to be wearable on the head or neck (e.g., forehead, ear, neck, nose, and / or tongue). In several embodiments, attenuation or blocking of nerve impulses and / or neurotransmitters is provided. In some embodiments, nerve impulses and / or neurotransmitters are enhanced. In some embodiments, the device is configured to be wearable on or proximal to the user's ear (e.g., auricular neuromodulation (e.g., neural stimulation) including, but not limited to, the auricular branches of the vagus nerve). The device can be monoauricular or binaural, including a single device or multiple devices connected wired or wirelessly.

[0062] When using non-invasive or wearable neuromodulation devices, there is a need for systems with a compact, ergonomic form factor to enhance efficacy, compliance, and / or comfort. In several embodiments, neuromodulation systems and methods are provided that enhance or inhibit nerve impulses and / or neurotransmission, and / or modulate the excitability of nerves, neurons, neural circuits, and / or other neuroanatomical sites that affect nerve and / or neuron activation. For example, neuromodulation (e.g., neural stimulation) can include one or more of the following effects on neural tissue: depolarizing a neuron such that the neuron fires an action potential; hyperpolarizing a neuron to inhibit an action potential; depleting neuronal ion stores to inhibit firing an action potential; changing with proprioceptive input; affecting muscle contraction; affecting changes in neurotransmitter release or uptake; and / or inhibiting firing.

[0063] In some embodiments, the wearable systems and methods disclosed herein can be advantageously used to identify whether a treatment is effective in significantly reducing or preventing a medical condition, including but not limited to tremor severity. Wearable sensors can advantageously monitor, characterize, and assist in the clinical management of hand tremors and other medical conditions (including those disclosed elsewhere herein). Without being bound by theory, the clinical assessment of a medical condition (e.g., tremor severity) can be related to wrist movement measurements taken simultaneously using an inertial measurement unit (IMU). For example, tremor characteristics extracted from an IMU at the wrist can provide information about the tremor phenotype, which can be used to improve diagnosis, prognosis, and / or treatment outcomes. Kinematic metrics can be related to tremor severity, and, for example, machine learning algorithms incorporated into the neuromodulation systems and methods disclosed herein can predict a visual assessment of tremor severity.

[0064] Neuromodulation device

[0065] Figure 1A A block diagram illustrating an exemplary neuromodulation (e.g., nerve stimulation) device 100 is shown. Device 100 includes a plurality of hardware components capable of or programmed to deliver therapy across a user's skin. As Figure 1A illustrated, some of these hardware components may be optional, as indicated by the dashed boxes. In some cases, device 100 may include only the hardware components required for the stimulation therapy. The hardware components will be described in more detail below.

[0066] Device 100 may include two or more effectors, e.g., electrodes 102 for delivering nerve stimulation signals. In some cases, device 100 is configured for percutaneous use only and does not include any percutaneous or implantable components. In some embodiments, the electrodes may be dry electrodes. In some embodiments, water or gel may be applied to the dry electrodes or the skin to improve conductivity. In some embodiments, the electrodes do not include any hydrogel materials, adhesives, or the like.

[0067] Device 100 may further include stimulation circuitry 104 for generating the signals applied through the electrode(s) 102. The frequency, phase, timing, amplitude, or offset of the signals may vary. Device 100 may also include power electronics 106 for providing power to the hardware components. For example, power electronics 106 may include a battery.

[0068] The device 100 may include one or more hardware processors 108. The hardware processor 108 may include a microcontroller, a digital signal processor, an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof designed to perform the functions described herein. In one embodiment, all processing discussed herein is performed by the hardware processor(s) 108. The memory 110 may store patient-specific data and the rules discussed below.

[0069] In the illustrated figures, the device 100 may include one or more sensors 112. As shown in the figures, the sensor(s) 112 may be optional. The sensors may include, for example, biomechanical sensors and / or bioelectrical sensors (e.g., EMG, EEG, and / or nerve conduction sensors) configured to measure movement, for example. The sensors may include, for example, cardiac activity sensors (e.g., ECG, PPG), skin conductance sensors (e.g., galvanic skin response, skin electrical activity), and motion sensors (e.g., accelerometers, gyroscopes). The one or more sensors 102 may include an inertial measurement unit (IMU).

[0070] In some embodiments, the IMU may include one or more of a gyroscope, an accelerometer, and a magnetometer. The IMU may be attached or integrated into a neuromodulation (e.g., nerve stimulation) device 100. In one embodiment, the IMU is an off-the-shelf component. In addition to its general meaning, the IMU may also include specific components discussed below. For example, the IMU may include yet another sensor capable of collecting motion data. In one embodiment, the IMU includes an accelerometer. In some embodiments, the IMU may include multiple accelerometers to determine motion on multiple axes. Additionally, in further embodiments, the IMU may also include one or more gyroscopes and / or magnetometers. Since the IMU may be integrated with the nerve stimulation device 100, the IMU may generate data from its sensors that respond to motion, movement, or vibration sensed by the device 100. Additionally, when the user wears the device 100 with an integrated IMU, the IMU may enable the detection of the user's voluntary and / or involuntary movements.

[0071] Device 100 may optionally include user interface components (such as feedback generator 114 and display 116). Display 116 may provide instructions or information related to calibration or therapy to the user. For example, display 116 may also provide alerts (such as an indication of the response to therapy). Feedback generator 114 may also be used to provide alerts, and the feedback generator may provide tactile feedback to the user (such as at the start or termination of stimulation, for prompting an alert to remind the user of a troubleshooting situation, to perform tremor-induced activities to measure tremor movement, etc.). Thus, user interface components such as feedback generator 114 and display 116 may provide audio, visual, and tactile feedback to the user.

[0072] In addition, device 100 may include communication hardware 118 for wireless or wired communication between device 100 and an external system (such as the user interface device discussed below). Communication hardware 118 may include an antenna. Communication hardware 118 may also include an Ethernet or data bus interface for wired communication.

[0073] Although the illustrated diagrams show several components of device 100, some of these components are optional and not required in all embodiments of device 100. In some embodiments, the system may include diagnostic devices or components that do not include a neuromodulation function. The diagnostic device may be a companion wearable device wirelessly connected via a connected cloud server and includes, for example, sensors (such as the heart activity, skin conductance, and / or motion sensors described elsewhere herein).

[0074] In some embodiments, device 100 may also be configured to deliver one, two, or more of the following: magnetic stimulation, vibration stimulation, mechanical stimulation, thermal stimulation, ultrasonic stimulation, or other forms of stimulation instead of or in addition to electrical stimulation. Such stimulation may be delivered via one, two, or more effectors in contact with or proximal to the skin surface of the patient. However, in some embodiments, the device is configured to deliver only electrical stimulation and is not configured to deliver one or more of magnetic stimulation, vibration stimulation, mechanical stimulation, thermal stimulation, ultrasonic stimulation, or other forms of stimulation.

[0075] Although several nerve stimulation devices are described herein, in some embodiments, the nerves are non-invasively regulated to achieve nerve inhibition. Nerve inhibition can occur in a variety of ways, including but not limited to hyperpolarizing the neurons to inhibit action potentials and / or depleting the neuronal ion stores to inhibit the firing of action potentials. In some embodiments, this can occur via, for example, anodic or cathodic stimulation, low-frequency stimulation (e.g., less than about 5 Hz in some cases), or continuous or intermediate burst stimulation (e.g., theta burst stimulation). In some embodiments, the wearable device has at least one implantable portion, which can be temporary or more long-term. In many embodiments, the device is fully wearable and non-implantable.

[0076] User interface device

[0077] Figure 1B Illustrated is the communication between the nerve stimulation device 100 and the user interface device 150 via the communication link 130. The communication link 130 can be wired or wireless. The nerve modulation (e.g., nerve stimulation) device 100 is capable of communicating and receiving instructions from the user interface device 150. The user interface device 150 can include a computing device. In some embodiments, the user interface device 150 is a mobile computing device (such as a mobile phone, smartwatch, tablet, or wearable computer). The user interface device 150 can also include a server computing system remote from the nerve stimulation device. The user interface device 150 can include a hardware processor(s) 152, a memory 154, a display 156, and power electronics 158. In some embodiments, the user interface device 150 can also include one or more sensors (such as the sensors described elsewhere herein). Additionally, in some cases, the user interface device 150 can generate an alert in response to a device problem or a response to a therapy. An alert from the nerve stimulation device 100 can be received.

[0078] In additional embodiments, the data obtained from the one or more sensors 102 is processed by a combination of the hardware processor(s) 108 and the hardware processor(s) 152. In further embodiments, the data collected from the one or more sensors 102 is transmitted to the user interface device 150 in cases where the hardware processor 108 performs little or no processing. In some embodiments, the user interface device 150 can include a remote server that processes the data and transmits signals back to the device 100 (e.g., via the cloud).

[0079] Figure 1CSchematically illustrates a neuromodulation device and a base station. The device can include a stimulator and a detachable band, which includes two or more working electrodes (positioned above the median nerve and the radial nerve) and a counter electrode positioned on the dorsal side of the wrist. The electrodes can be, for example, dry electrodes or hydrogel electrodes. The base station can be configured to periodically (e.g., daily) stream movement sensor and usage data and charge the device. The device stimulation burst frequency can be calibrated for a lateral posture holding task “wing-beating” or a forward posture holding task for a predetermined time (e.g., 20 seconds for each subject). Other non-limiting examples of device parameters can be as disclosed elsewhere herein.

[0080] In some embodiments, the stimulation can alternate between each nerve such that the nerves are not stimulated simultaneously. In some embodiments, all the nerves are stimulated simultaneously. In some embodiments, the stimulation is delivered to each nerve in one of a number of burst patterns. Stimulation parameters can include on / off, duration, intensity, pulse rate, pulse width, waveform shape, and the slopes of the on and off of the pulses. In a preferred embodiment, the pulse rate can be from about 1 to about 5000 Hz, about 1 Hz to about 500 Hz, about 5 Hz to about 50 Hz, about 50 Hz to about 300 Hz, or about 150 Hz. In some embodiments, the pulse rate can be from 1 kHz to 20 kHz. The preferred pulse width can be in the range of 50 to 500 μs (microseconds) in some cases (such as about 300 μs). The intensity of the electrical stimulation can vary from 0 mA to 500 mA, and in some cases, the current may be about 1 to 11 mA. The electrical stimulation can be adjusted in different patients and with different electrical stimulation methods. The increment of intensity adjustment can be, for example, from 0.1 mA to 1.0 mA. In a preferred embodiment, the stimulation can last for about 10 minutes to 1 hour (such as about 10, 20, 30, 40, 50, or 60 minutes), or a range including any two of the foregoing values. In some embodiments, multiple electrical stimulations can be delivered offset from each other in time by a predetermined fraction that is a multiple of the period of a measured rhythmic biological signal (such as hand tremor) (e.g., about 1 / 4, 1 / 2, or 3 / 4 of the period of the measured signal). Further possible stimulation parameters are described, for example, in U.S. Patent 9,452,287 to Rosenbluth et al., U.S. Patent No. 9,802,041 to Wong et al., PCT Publication No. WO 2016 / 201366 to Wong et al., PCT Publication No. WO 2017 / 132067 to Wong et al., PCT Publication No. WO 2017 / 023864 to Hamner et al., PCT Publication No. WO 2017 / 053847 to Hamner et al., PCT Publication No. WO 2018 / 009680 to Wong et al., and PCT Publication No. WO 2018 / 039458 to Rosenbluth et al., each of the foregoing being incorporated herein by reference in its entirety.

[0081] Controller

[0082] Figure 2 A block diagram illustrating an embodiment of the controller 200, which can be used with the above regarding Figure 1A - Figure 1Cimplemented by the described hardware components. The controller 200 may include multiple engines for performing the processes and functions described herein. The engines may include programming instructions for performing the processes discussed herein for detecting input conditions and controlling output conditions. The engines may be executed by the one or more hardware processors of the neuromodulation (e.g., neural stimulation) device 100 alone or in combination with the patient monitor 150. The programming instructions may be stored in the memory 110. The programming instructions may be implemented using C, C++, JAVA, or any other suitable programming language. In some embodiments, some or all of the portions of the controller 200 that include the engines may be implemented in dedicated circuitry such as ASICs and FPGAs. Some aspects of the functions of the controller 200 may be executed remotely on a server (not shown) over a network. Although shown as separate engines, the functions of the engines discussed below do not necessarily need to be separate. Thus, the controller 200 may be implemented with the hardware components described above with respect to Figure 1A - Figure 1C as described.

[0083] The controller 200 may include a signal collection engine 202. The signal collection engine 202 may enable the acquisition of raw data (including but not limited to accelerometer or gyroscope data from the IMU 102) from sensors embedded in the device. In some embodiments, the signal collection engine 202 may also perform signal preprocessing on the raw data. The signal preprocessing may include noise filtering, smoothing, averaging, and other signal preprocessing techniques to clean the raw data. In some embodiments, a portion of the signal may be discarded by the signal collection engine 202.

[0084] The controller 200 may also include a feature extraction engine 204. The feature extraction engine 204 may extract relevant features from the signals collected by the signal collection engine 202. The features may be in the time domain and / or frequency domain. For example, some of the features may include amplitude, bandwidth, area under the curve (e.g., power), energy in frequency bins, peak frequency, ratio between frequency bands, etc. Signal processing techniques such as Fourier transform, bandpass filtering, lowpass filtering, highpass filtering, etc. may be used to extract the features.

[0085] The controller may further include a rule generation engine 206. The rule generation engine 206 may use the extracted features from the collected signals and determine rules corresponding to the neuromodulation therapy. The rule generation engine 206 may automatically determine the correlation between the specifically extracted features and the neuromodulation therapy outcome. The outcomes may include, for example, identifying patients who will respond to the therapy (e.g., during an initial trial fitting or calibration process) based on tremor characteristics (e.g., approximate entropy) from kinematic data, predicting the stimulation settings that will achieve the best therapeutic effect (e.g., dose, where the dose of the treatment or the parameters of the administration include, but are not limited to, the duration of the stimulation, the frequency and / or amplitude of the stimulation waveform, and the daily stimulation time applied) for a given patient (based on their tremor characteristics), predicting the patient's tremor severity at a given point, predicting the patient's response over time, examining the patient's medication responsiveness in combination with tremor severity over time, predicting the response to percutaneous or transcutaneous stimulation, or implantable deep brain stimulation or thalamotomy based on tremor characteristics and severity over time, and predicting the optimal time for a patient to receive percutaneous or transcutaneous stimulation, or deep brain stimulation or thalamotomy based on tremor characteristics and severity over time, predicting the patient-reported therapy outcome or patient-reported satisfaction using tremor characteristics evaluated from kinematic measurements of the device; predicting the patient's response to an undesired user experience using tremor characteristics evaluated from kinematic measurements of the device and the patient usage log, where the undesired user experience may include, but is not limited to, device failure and adverse events such as skin irritation or burns; predicting the patient response trend based on tremor severity, where the trend may be evaluated across the total number of segments within a single patient, or across a patient population; predicting or classifying subtypes of tremor based on kinematic analysis of tremor characteristics to predict patient response; predicting or classifying subtypes of tremor to provide guidance for optimized therapy parameters for an individual; predicting or classifying subtypes of tremor to optimize future study designs based on the subtype (e.g., selecting a specific subtype of essential tremor for a clinical study with a specific design that seeks to address the therapy needs for that subtype); and predicting the patient or customer satisfaction (e.g., net promoter score) based on the patient's response or other kinematic features from measuring tremor movement. In some embodiments, the essential tremor pathology may include, for example, a major cerebellar variant with Bergmann gliosis and Purkinje cell torpedoes, and Lewy body variants, and dystonia variants, and multiple sclerosis variants, and Parkinson's disease variants.

[0086] In some embodiments, a neuromodulation (e.g., nerve stimulation) device may apply percutaneous stimulation to a patient with tremors who is a candidate for implantable deep brain stimulation or thalamotomy. Other sensor measurements of tremor characteristics and tremor severity will be used to evaluate the response during a pre-specified usage period, which may be 1 month or 3 months, or 5, 7, 14, 30, 60 or 90 days or more or less. The response to percutaneous stimulation, evaluated using sensor measurements from the device, for example by the algorithms described herein, can advantageously provide input to a predictive model that provides an assessment of the likelihood that the patient will respond to implantable deep brain stimulation or other implantable or non-implantable therapies.

[0087] In some embodiments, when a tremor induction task is being performed, a neuromodulation (e.g., nerve stimulation) device or a second device with sensors may collect motion data or data from other sensors. The patient may be directly instructed to perform the task, for example via a display or audio on the device. In some embodiments, the characteristics of the tremor induction task are stored on the device and are used to automatically activate the sensors to measure data and store the data to memory during the relevant tremor task. The time period for measuring and storing data may be, for example, 10, 20, 30, 60, 90, 120 seconds, or 1, 2, 3, 5, 10, 15, 20, 30 minutes, or 1, 2, 3, 4, 5, 6, 7, 8 hours, or more or less, or a range incorporating any two of the foregoing values. Based on a training data set from a cohort of wearers who previously had tremors or other conditions, a feature extraction engine may detect features related to the response to the stimulation such that a quantitative and / or qualitative likelihood that the patient will respond or not respond to treatment can be presented to the patient or the doctor. This data may be measured in some cases before a neuromodulation (e.g., nerve stimulation) prescription is issued or during a trial period. In another embodiment, the features may be related to the type of tremor measured (such as resting tremor (associated with Parkinson's disease), postural tremor, kinetic tremor, intention tremor, rhythmic tremor (e.g., a single dominant frequency), or a mixed tremor (e.g., multiple frequencies)). The most likely type of tremor to be detected may be presented to the patient or the doctor as a diagnosis or information assessment before receiving the stimulation, or to evaluate the appropriateness of issuing a neuromodulation (e.g., stimulation treatment) prescription. In another embodiment, various stimulation patterns may be applied based on the determined tremor type; different patterns may be applied, and the different patterns may include variations in stimulation parameters such as frequency, pulse width, amplitude, burst frequency, stimulation duration, or daily stimulation time. In one embodiment of a smartphone, tablet, or other device, the task of inducing tremors may be included in an application that requires the patient to take a photo of themselves, which causes the patient to perform tasks with postures and intentional movements.

[0088] In some embodiments, a neuromodulation (e.g., nerve stimulation) device or a second device with sensors can collect motion data or data from other sensors, and can measure data over longer time periods (e.g., 1, 2, 3, 4, 5, 10, 20, 30 weeks, 1, 2, 3, 6, 9, 12 months, or 1, 2, 3, 5, 10 years, or more or less, or a range incorporating any two of the foregoing values) to determine characteristics or biomarkers associated with the onset of a tremor disorder (such as essential tremor, Parkinson's disease, dystonia, multiple sclerosis, etc.). Biomarkers can include specific changes over time of one or more characteristics of the data, or one or more characteristics crossing a predefined threshold. In some embodiments, characteristics of tremor-inducing tasks have been stored on the device and are used to automatically activate the sensors when those tremor-inducing tasks are performed, to measure data during relevant time periods and store the data in memory.

[0089] In some embodiments, the rule generation engine 206 relies on calibration instructions to determine rules between characteristics and outcomes. The rule generation engine 206 can employ machine learning modeling as well as signal processing techniques to determine rules, where the machine learning modeling and signal processing techniques include, but are not limited to: supervised and unsupervised algorithms for regression and classification. Specific types of algorithms include, for example, artificial neural networks (perceptron, backpropagation, convolutional neural network, recurrent neural network, long short-term memory network, deep belief network), Bayesian (naive Bayes, multinomial Bayes, and Bayesian network), clustering (k-means, expectation maximization, and hierarchical clustering), ensemble methods (classification and regression tree variants and boosting algorithms), instance-based (k-nearest neighbor, self-organizing map, and support vector machine), regularization (elastic net, ridge regression, and least absolute shrinkage and selection operator), and dimensionality reduction (principal component analysis variables, multidimensional scaling, discriminant analysis variables, and factor analysis). In some embodiments, the controller 200 can use the rules to automatically determine outcomes. The controller 200 can also use the rules to control or change the settings of the nerve stimulation device, including but not limited to stimulation parameters (e.g., stimulation amplitude, frequency, patterning (e.g., burst stimulation), interval, daily time, individual segments, or cumulative time, etc.).

[0090] Accordingly, the rules can improve the operation of a neuromodulation (e.g., nerve stimulation) device and advantageously and accurately identify potential candidate therapies over time as well as various disease states and therapy parameters. The generated rules can be stored in memory 110 and / or memory 154. For example, the rules can be generated after calibration and stored prior to operation of the nerve stimulation device 100. Thus, in some embodiments, the rules application engine 208 can apply the stored rules to new data collected by the IMU to determine results or control the neuromodulation (e.g., nerve stimulation) device 100.

[0091] Specific examples of calibration and determination of the rules will be described in more detail below.

[0092] Calibration

[0093] In some embodiments, a neuromodulation device can include the ability to track a user's motion data for measuring one, two, or more tremor frequencies of a patient. The patient may have a single tremor frequency or, in some cases, exhibit multiple discrete tremor frequencies when performing different tasks. Once the tremor frequencies are observed, they can be used as one of many pioneering input parameters for customizing neuromodulation therapy. The therapy can be delivered, for example, percutaneously via one, two, or more nerves (e.g., the median and radial nerves, and / or other nerves disclosed elsewhere herein) to alleviate or improve the patient's condition (including but not limited to their tremor burden). In some embodiments, the therapy modulates afferent nerves but not efferent nerves. In some embodiments, the therapy preferentially modulates afferent nerves. In some embodiments, the treatment does not involve functional electrical stimulation. The tremor frequency can be used to calibrate a patient's neuromodulation therapy and, in some embodiments, is used as a calibration frequency to set one or more parameters of the neuromodulation therapy (e.g., burst envelope period). In some embodiments, the calibration frequency can be between, for example, about 4 Hz and about 12 Hz, between about 3 Hz and about 6 Hz, or about 3 Hz, 4 Hz, 5 Hz, 6 Hz, 7 Hz, 8 Hz, 9 Hz, 10 Hz, 11 Hz, or 12 Hz, or a range including any two of the foregoing values.

[0094] In some embodiments, a hardware processor can be configured to perform any number of the following: obtain raw motion data by turning on an IMU transducer (e.g., an accelerometer); and collect patient data over a selected time period (e.g., 10 seconds) (x / y / z axes); perform a fast Fourier transform (FFT) on the x-axis (resulting in ); perform an FFT on the y-axis (resulting in ); and / or perform an FFT on the z-axis (resulting in ); And determine the calibration frequency as the maximum value of the TFD (e.g., in some cases, between about 4 Hz and about 12 Hz).

[0095] Figure 3 Illustrates an embodiment of a process 300 for calibrating a neuromodulation (e.g., nerve stimulation) device 100. Process 300 can be implemented by any of the systems discussed above. Process 300 can be implemented to calibrate a specific neuromodulation (e.g., nerve stimulation) device 100 for a particular user or multiple nerve stimulation devices across multiple users.

[0096] In one embodiment, when the neuromodulation (e.g., nerve stimulation) device 100 is activated, the calibration process 300 begins at block 302. The device can be activated in response to a user input. The user input can be received via a button or any other user interface (e.g., the display 106 of the neuromodulation (e.g., nerve stimulation) device). In some embodiments, the neuromodulation (e.g., nerve stimulation) device 100 can be activated based on a signal received from a patient monitor 150 or another computing system.

[0097] After activating the calibration, the controller 200 can begin collecting sensor data from the IMU 102. In one embodiment, the controller 200 can continue to collect sensor data for a period of 10 seconds, or about, at least about, or no more than about 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 20, 25, 30, 40, 50, 60 seconds, including ranges of any two of the foregoing values, or other time periods. In some embodiments, the controller can continue to collect data for a longer period (e.g., 1, 2, 3, 4, 5, 6, 7, 8 hours, or 12 or 24 hours, 1, 2, 3, 4 weeks, 1, 2, 3, 4, 5, 6 months or longer, or indefinitely). After the data collection period, at block 306, the controller 200 can activate the electrodes 104 to apply nerve stimulation. At block 308, after applying the nerve stimulation for a period of time, the controller 200 can obtain additional data from the IMU 102. In one embodiment, the post-stimulation data collection period is 10 seconds.

[0098] At block 310, the controller 200 can process the collected sensor data before and after the stimulation to determine one or more of the results discussed above (including determining one or more rules).

[0099] In some cases, the calibration segment can be processing- and / or data-intensive. In particular, when performing calibration on a device, the on-board hardware processor and memory may not have sufficient processing or storage capabilities to handle large continuous data sets. For example, assuming a sampling frequency of 104 Hz, a 10-second calibration can generate 1,040 samples for each axis. Additionally, each sample may require 4 bytes of data storage to store floating-point data. Thus, the total data storage may correspond to 1,040 * 3 * 4 = 12.48 KB. To reduce memory usage, it may be advantageous to analyze the calibration signal in an analysis window in some cases. An embedded processing device can more efficiently analyze samples (which are powers of 2). Thus, in some cases, 256 samples are selected, which corresponds to 2.46 seconds at a sampling rate of 104 samples per second. Other sampling rates can also be used. The corresponding data usage is 256 * 3 * 4 = 3.072 KB, which is significantly less than 12.48 KB. Note that the Nyquist frequency is 52 Hz, resulting in a frequency bin resolution of 0.40625 Hz. Windowed calibration can advantageously reduce the memory bandwidth required within the neuromodulation device, potentially allowing for increased speed and real-time or near-real-time processing.

[0100] Figure 3A A flowchart illustrating a non-limiting implementation of a process for performing windowed calibration is shown. In some implementations, the process can include any of the following: collecting multiple windows of x, y, z, accelerometer data (such as (5) 2.4-second windows); performing an FFT on each window; summing A + B + C for each window; averaging all (e.g., 5) FFTs; and performing peak detection. Thus, the processing is performed on the windows and then combined as shown in Figure 3A to reduce memory usage. In some cases, certain portions of the signal or window affected by artifacts or noise can also be discarded (as described herein) to reduce processing requirements. Sampling over a shorter time period (such as a window) as compared to a long time period may provide tremor frequency variability on a single capture. In some cases, tremor frequency variability is an indicator for determining the therapeutic effect of a stimulation, where lower variability indicates a higher likelihood of a successful treatment outcome. Tremor frequency variability can be determined as the change in frequency at which tremor peaks occur, the change in power around the tremor peak frequency (as described herein), or a combination of these metrics and any other metric related to the nature of the tremor peak. Calibration can also vary based on the task, resulting in different calibration frequencies for a particular task. In some cases, the calibration can automatically determine the type of task the user is performing.

[0101] In some embodiments, the hardware processor may be configured to diversify calibration activities and / or update calibration on a section-by-section basis. Peak tremor frequencies may be captured from kinematic measurements during different types of activities or tasks, such as 2, 3, 4, 5, 6, 7, 8, 9, 10 or more tasks. In some cases, a patient may have a first tremor frequency when performing a first activity or task, a second tremor frequency when performing a second activity or task, and / or a third tremor frequency when performing a third activity or task, and so on. In some cases, these tremor frequencies may differ by up to 2.5 Hz or more (e.g., about 0.5 Hz, 1 Hz, 1.5 Hz, 2 Hz, 2.5 Hz, 3 Hz, or more or less, or a range including any two of the foregoing values). Figure 3B is a figure that illustrates that, in some cases, a patient's tremor frequency may vary across different tasks, including holding, movement, a first drawing A task, a second drawing B task, and tipping relative to a postural holding task. Without being limited by theory, when neuromodulation (e.g., nerve stimulation) alternates between the first and second nerves (such as the median nerve and the radial nerve) at the tremor frequency for a given task, treatment outcomes may be improved. Thus, the alternating frequency may be modified based on the task performed before, during, or after stimulation to improve the patient's response. For example, this may involve detecting different types of activities moment by moment during stimulation; extracting the tremor frequency moment by moment during a specific activity identified by the patient or by an algorithm based on the unique kinematic features of the tremor-induced task; and / or measuring the tremor frequency during tremor tasks of posture, movement, and locomotion, storing the frequency for each task in the memory of the device, and allowing the patient to select the desired task type or modify based on movement during neuromodulation (e.g., nerve stimulation). For example, in some embodiments, the neuromodulation device may be configured to use a first calibration frequency for a first stimulation therapy during a first activity and a second different calibration frequency for a second stimulation therapy during a second different activity.

[0102] Data collection

[0103] Figure 4 illustrates a flowchart of an embodiment of a process 400 for collecting data from the IMU 102 and processing the collected data. The process 400 may be implemented by any of the above systems.

[0104] In one embodiment, process 400 begins at block 402 when controller 200 receives sensor data from IMU 102 in the time domain. The sensor data may include raw accelerometer and / or gyroscope data, and / or other sensor data. In one embodiment, the sensor data may include data from each of the three axes. At block 404, controller 200 may separate the received sensor data into four non-overlapping frames, each having a length of approximately 2.5 seconds or other desired time period. At block 406, controller 406 may transform each frame into the frequency domain. At block 408, controller 200 may combine the resulting spectra after frequency transformation from each axis frame. In one embodiment, the controller combines these spectra based on user-specified input. The magnitude spectrum may be calculated using the discrete-time Fourier transform. Further, the power spectral density (PSD) estimate may be calculated using the periodogram with a Hann window. In one embodiment, the L1 or L2 norm spanning each frequency bin may be used to combine the spectra from each of the three axes for a given frame:

[0105] L1: P c，f = |P x，f + P y，f + P z，f |

[0106] L2:

[0107] where P(x, f) is the magnitude or power in frequency bin f of channel x.

[0108] Different combinations of spectral features can be used for classifying and predicting tremor characteristics. For example, frequency components from approximately 0 Hz to approximately 2.5 Hz can be used to detect periods of non-tremor movement, while the frequency range between approximately 4 and approximately 12 Hz can be used to detect features related to tremors, or approximately 3 to approximately 8 Hz for different tremor variants (including but not limited to Parkinson's disease). Kinematic features of the raw data signal from the IMU, such as the regularity, magnitude, and shape of the signal, can be used for classification.

[0109] At blocks 410 - 416, the controller 200 can examine the combined spectrum of each of the four frames for viewing non - tremor movement artifacts. In one embodiment, at block 410, the controller 200 calculates the total power in a low - frequency band. The low - frequency band can include frequencies, for example, between approximately 0 Hz and 2.5 Hz. These frequencies can be specified as an input by the user. Further, at block 412, the controller 200 calculates the energy in a second frequency band that is higher than the low - frequency band. In one embodiment, the second frequency band includes frequencies between 4 Hz and 12 Hz, which includes typical tremor frequencies and generally 2.75 Hz around the peak tremor frequency. In one embodiment, the controller 200 searches for tremor spikes in the second frequency band. The user can specify the second frequency band as an input.

[0110] At block 414, the controller 200 can compare the total power in the first frequency band and the second frequency band. The comparison can include dividing the total power in the second frequency band by the first frequency band. At block 416, the controller 200 can compare this ratio with a pre - determined threshold to determine whether a particular frame can be used for further analysis. In one embodiment, if the controller 200 determines that the total power of the frame in the second frequency band is greater than the total power in the first frequency band, or both the total tremor frequency band and the total low - frequency band are less than a heuristically determined threshold, then the controller 200 can determine that the frame has no non - tremor movement artifacts. In some cases, the threshold can be input by the user. Additionally, in some cases, the threshold can be approximately 0.8. The threshold can depend on system or device parameters. In some cases, the threshold is from approximately 0.5 to approximately 0.9. Frames identified as having no movement or no artifacts can be used for further analysis. Frames with artifacts can be discarded or ignored by the controller 200. At block 416, the controller 200 can combine the frames without artifacts. In one embodiment, the controller 200 can average the remaining frames to produce a single spectrum.

[0111] At block 418, the controller 200 may use the processed data to determine metrics that may be related to the results of the neuromodulation therapy. In one embodiment, the controller 200 may examine acute (segment time frame) and chronic (overall treatment) effects. The controller may extract different metrics from the combined spectrum of no movement recorded before and after each stimulation. These metrics include the amplitude / PSD at the peak tremor frequency (which, in some cases for Parkinson's tremor, is generally limited between 4 - 12 Hz, or between 3 - 8 Hz), the total amplitude / PSD in a frequency band around the peak tremor frequency (generally 2.75 Hz wide), the total amplitude / PSD in a wide frequency band (generally 4 - 12 Hz), or the total amplitude / PSD in a frequency band around the peak tremor frequency selected only from the pre-stimulation spectrum. Parameters such as bandwidth and search frequency may be predetermined or received as input from the user. In some cases, hand tremors in Parkinson's disease are severe during rest (in contrast, essential tremors are severe during movement), such as when the hand is placed on the thigh or while walking. Tremors measured during walking may involve different rules for extracting the tremor signal and analysis. The tremor induction tasks for Parkinson's disease therapy may be different from those for essential tremors. Non-limiting examples of tremor induction tasks for Parkinson's disease may include, for example, placing the hand on the thigh or on a table (e.g., when the hand, arm, or leg muscles are relaxed).

[0112] Figure 4A A flowchart illustrating an embodiment of process 401 for collecting data from the IMU 102 and processing the collected data. Process 401 may be implemented by any of the systems described above. Process 401 may include Figure 4 any number of blocks of process 400, and also includes additional blocks as part of determining the frames for analysis in block 416 or in addition to determining the frames for analysis in block 416. For example, process 401 may include collecting sensor data in the time domain at block 416A. The data may relate to, for example, any number of approximate entropy, displacement, metrics for summarization of continuous data and / or functional data analysis (e.g., curve fitting, functional PCA, filtering, mean, median, range, etc.), and / or event detection using supervised and unsupervised methods (e.g., movement classification). At block 416B, the collected data may be separated into frames. At block 416C, metrics (e.g., non-spectral analysis) may be calculated based on all or a subset of the frames of the time series data. For example, in the case of approximate entropy, the frames to be used for calculation may be determined based on the intermediate results of the PSD analysis stream (schematically shown as line 416').

[0113] As regarding Figure 3As discussed, when using the above signal collection and preprocessing during calibration, the controller 200 can check certain exclusion criteria to determine whether the metrics from the section can be used in subsequent analysis. These analyses can be performed by algorithms, such as by comparing and power associated with predefined spectra known to be associated with voluntary and tremor movements. Characteristics of the raw signal (i.e., in the time domain) or the frequency domain transformed version can be used to generate new rules to classify whether a frame of data should be included or excluded. These criteria can include at least one frame in the pre-stimulus and post-stimulus recordings (which has no movement artifacts), at least one frame in the pre-stimulus and post-stimulus recordings (which contains tremors), the shortest time between the end of the stimulus and the post-stimulus recording, and the shortest time between the pre-stimulus recording and the previous stimulation section.

[0114] Rule generation

[0115] Figure 5A FIG. 5 illustrates a flowchart of an implementation of a process 500 for generating rules for determining the outcome of a nerve stimulation therapy. The process 500 can be implemented by any of the above systems. As discussed above with respect to Figure 4 the controller 200 can collect and process data from the IMU 402 (and / or other sensors). The controller 200 can extract features or metrics from the processed data. In some embodiments, the controller 200 does not perform motion artifact removal as discussed above with respect to blocks 410-416. Thus, the controller 200 can use all frames to generate rules.

[0116] The process 500 can start at block 502, where the controller 200 can extract features from the collected signals. The features can include time domain features and / or frequency domain features. The features can be pre-determined. For example, the controller 200 can search for certain features in the received signal or the processed signal. At block 504, the controller 200 can correlate the extracted features with a particular application or the outcome of a nerve stimulation therapy. The correlation can include machine learning algorithms. Based on the correlation, the controller 506 can generate rules at block 506. In some embodiments, the controller 200 determines the rules based on training set data collected from many patients. In other embodiments, the controller 200 determines the rules based on calibration data received from a particular patient. Additionally, the controller 200 can update previously determined rules based on additional data processed while using the nerve stimulation device 100.

[0117] The following is an example of the rule generation process 500. In one embodiment, the controller 200 obtained an accelerometer signal from the user's dominant hand. The signal was divided into 2.5-second frames (50% overlap), and the amplitude spectrum was calculated for each axis frame and used the above with respect to Figure 4The described L1-norm method is combined. In this particular example, since the experiment is not self-orienting, motion artifact detection is not applied. As Figure 5B shown in Figure 5B , several different features are extracted from the combined spectrum of each frame. The controller 200 can combine these features across all frames from a given recording. The combination can include averaging the values from each frame and finding the minimum and maximum values across all frames. In one embodiment, all amplitude-spectrum-based features are log-transformed by the controller 200. Additionally, the controller 200 also uses the duration of the recording and the type of task as features. For example, the controller 200 can receive the type of task (as an input).

[0118] Using these features, the controller 200 creates a random forest classifier to predict the severity of tremors, as measured by a clinical score that assigns a value between 0 and 4 to each task, where 4 is the maximum impairment. Out of approximately 1000 recordings, 100 were randomly selected that had a clinical score distribution similar to the rest of the data and were reserved as a test set. For the rest of the data, 5-fold cross-validation and random search were used to optimize the random forest hyperparameters. The optimized hyperparameters include the number of decision trees, the maximum features used to split nodes, the maximum tree depth, the minimum data examples used to split nodes, and the minimum samples used to create leaves. The search for random combinations of hyperparameters was performed for 50 iterations, and the combination that produced the best cross-validation accuracy was selected. After selecting the best model hyperparameters, the model was retrained using all data except the test set and then evaluated using the reserved data. The algorithm can be developed with a training set of data from kinematics based on postural holding and kinematics based on controlled setting tasks and tested in kinematics based on real-world time-locked tasks and in dynamic kinematics.

[0119] Rule imposition

[0120] Rules can be stored in several ways, including but not limited to any number of the following: (1) After training on a cohort of data, the rules can be stored in the cloud. The data will be transmitted periodically (e.g., nightly), and once the data is transmitted, the rules are applied. After execution in the cloud, changes to the stimulus or outcome can be sent back to the device or patient monitor; (2) The rules can be stored in the memory of the device or patient monitor and executed on the processor. The collected data can be processed, and the rules can be applied either real-time after measurement or after a stimulus; and / or (3) After each therapy session, generation (and modification) of the rules can occur based on an assessment of tremor improvement and related features measured before, during, and after each stimulation session.

[0121] Additional applications and working examples

[0122] Non-invasive electrical stimulation of the peripheral nerves (e.g., in the wrist, including the median nerve and the radial nerve) can lead to a reduction in hand tremors in patients with essential tremor (ET). Motion sensors placed on the wrist can quantify the tremors and provide kinematic measures related to the clinical assessment of tremor severity. During the clinical assessment of hand tremors at different time points (e.g., up to 60 minutes or more) before and after non-invasive median and radial nerve stimulation in a single session, the severity of the tremors can be quantified using measurements from wearable wrist sensors.

[0123] Fifteen participants performed hand tremor-specific tasks (including the task of reaching from the finger to the nose (referred to as the action), as well as posture, drawing, and pouring tasks) according to the Fahn-Tolosa-Marin Clinical Rating Scale (FTM-CRS). During kinematic measurements while the wrist of the treated hand was moving, the tremor severity was visually assessed according to the FTM-CRS. Improvements in both the FTM-CRS and kinematic measurements showed that non-invasive stimulation of the median and radial nerves at the wrist significantly improved symptomatic hand tremors within 60 minutes after the end of the stimulation. It was then found that the measured kinematics were directly related to the visual clinical assessment, and subsequently, the measured kinematics were used to predict the clinical assessment through a supervised machine learning regression model. The model could predict the clinical assessment within ±0.62 or better of the visual clinical assessment unit on average, including but not limited to the posture holding task. The observed prediction error was less than the resolution of the clinical rating scale (1 unit), and it was as accurate as or better than that of a human rater. The combination of non-invasive neuromodulation and kinematic measurements of tremors immediately before and after the therapy session can be used to advantageously treat and automatically classify tremor reduction according to the stimulation in both clinical and home settings.

[0124] The learning program can include one or more individual in-office sessions. Three electrodes (e.g., hydrogel electrodes (or dry electrodes in other embodiments)) are positioned transcutaneously to target the median and radial nerves of each participant, including being placed circumferentially on the device housing (such as a detachable strap ( Figure 6A)) The first (e.g., median), second (e.g., radial), and third (e.g., ulnar) electrodes. The active leads are placed on the median and radial nerves on the palm and back of the wrist and are connected to a stimulator (in some embodiments, the stimulator may be integrated into the neuromodulation device or be movable or otherwise attached to the neuromodulation device). The ulnar electrode is connected to the back of the wrist. Using an IMU worn on the participant's wrist, the tremor frequency was measured during a specific time (e.g., 10 seconds) when the participant extended the arm forward in a forward posture. The stimulation amplitude was increased by a desired amount (e.g., 0.25 mA steps) until the participant reported paresthesia corresponding to the distribution of the median and radial nerves. The stimulation included a series of charge-balanced biphasic pulses, 300 μs per phase, with a 50 μs period between the two phases, delivered at a 150 Hz frequency. The stimulation alternated between the median and radial nerves at a frequency equal to the tremor frequency of each participant ( Figure 6 B). The final stimulation amplitude was selected to be below the highest stimulation level that caused muscle contraction and was comfortable for the participant. The stimulation period included 40 minutes of continuous stimulation at the selected amplitude.

[0125] The kinematics of the wrist were measured when the participant performed different tasks at multiple time points (e.g., 4 time points): baseline (pre), 20 minutes during stimulation (during), immediately after stimulation (post 0), 30 minutes after stimulation (post 30), and 60 minutes after stimulation (post 60) ( Figure 6 C). Multiple kinematic features were extracted from the accelerometer for machine learning algorithm development to quantify tremor power and other tremor characteristics. To quantify tremor power, the average power was calculated in the frequency band corresponding to the strongest tremor oscillation in the accelerometer signal (e.g., peak power frequency ±1.5 Hz) in the range of 4 - 12 Hz, for example. For each participant, the sensor expressed the tremor power measured at each time point as a logarithmic ratio relative to the tremor power measured in the pre-stimulation recording (Pre), which means that a decrease in tremor power results in a negative logarithmic ratio. This transformation was performed to compare tremor reduction across all tasks without considering the original tremor amplitude in each task. The first and last seconds of each kinematic record were discarded to exclude signals from the preparation for movement. The spectral analysis of the accelerometer data was performed using the Welch method (e.g., 2-second window, 50% overlap) to generate the power spectral density of the signal. The following features were extracted from the power spectral density plot ( Figure 6 D). The kinematic features in the time domain and frequency domain can be used as model input features. The peak tremor frequency (frequency 峰值 ) was defined as the frequency in the 4 - 12 Hz band with the highest power. The peak power (power 峰值) is defined as the power at the peak of the frequency having the highest power in the 4 - 12 Hz frequency band. Average peak power (Power 峰值平均值 ) is defined as the average power in the ±1.5 Hz window centered at the frequency having the highest power in the 4 - 12 Hz frequency band. Total peak power (Power 峰值合计 ) is the total power of the power in the ±1.5 Hz window centered at the frequency having the highest power in the 4 - 12 Hz frequency band. Total tremor band power (Power 4-12Hz合计 ) is the total power in the 4 - 12 Hz frequency band. Spectral entropy (Entropy 频谱 ) is the entropy of the power spectral density in the range of 4 - 12 Hz. The Q - factor is determined as the peak frequency divided by the frequency range where the spectral power is 50% higher than the peak power. Approximate entropy (Entropy 近似值 ) was calculated as a metric on the raw signal for two consecutive frames (5 seconds) of the maximum tremor axis for Pre using the accelerometer data and the matching axis for Post, to quantify the amount of temporal regularity of the time - series data. The running comparison length was 2, and the threshold was 0.2x the standard deviation of the data.

[0126] The extracted features of the kinematic data were used to train a supervised machine - learning model. The gradient - boosting tree model was selected because of its robustness in regression and classification applications. The model was implemented in Python using XGBoost. First, the dataset was randomly divided into a training set and a test set. To determine the impact of the training set size, the data ratios allocated to the training set and the test set were systematically varied (from 20% to 80%), and the prediction was repeated the desired number of times (e.g., 250 times for each allocation group). Stratification was used to maintain the distribution of the results in the training set and the test set. The training objective function was to minimize the squared error using an L2 regularization term.

[0127] To identify the optimal values of the model parameters, a grid search was performed on these parameters: the number of trees (e.g., from 5 to 55 in steps of 10), the depth of the trees (e.g., from 3 to 6 in steps of 1), and the learning rate (e.g., 0.0001, 0.002, 0.004, and 0.0008). The default values in XGBoost were not used for other model parameters. The accuracy of the candidate models in the grid - search parameter space was evaluated using 3 - fold cross - validation. The parameter set that produced the highest training accuracy was used to build the model to predict the clinical assessment of the test dataset.

[0128] The accuracy of the test prediction can be the mean absolute error (MAE) between the true and predicted clinical assessments. It was first calculated for the segments within each clinical assessment group (e.g., 0 to 4), and then averaged across all clinical assessment groups to account for the differences in the number of data points in each group.

[0129] To determine the contribution of different kinematic features to prediction, the feature importance of the model was examined. Feature gain was used as a measure of importance, which reflects the improvement in classification of the feature of interest on each branch of the decision tree. The average feature importance was calculated across 250 models for each training / test group. Statistical analysis was performed using the Python packages SciPy and NumPy, and visualization was performed using Matplotlib. Non-parametric comparison (Wilcoxon signed-rank test) and Pearson correlation method were used.

[0130] Table 1 below illustrates examples of tremor power changes in logarithmic and equivalent % units (median and 95% confidence intervals) for each task, with Wicoxon signed-rank test *p < 0.05 and **p < 0.01.

[0131] Table 1

[0132]

[0133] A decrease in sensor-measured tremor kinematics (28 - 59% improvement across tasks) was observed during the period immediately after stimulation (34 - 75% improvement) up to 60 minutes after stimulation (21 - 60% improvement). A sustained and significant negative log ratio was observed at the time points during, immediately after, 30 minutes after, and 60 minutes relative to the pre-stimulation segment ( Figure 7 A - Figure 7 D, Table 1). The time course in the action task showed a significant decrease at all time points within one hour ( Figure 7 A, Table 1). The time course of plotting (e.g., spiral plotting) showed a significant decrease in tremor within 30 minutes after stimulus offset ( Figure 7 B, Table 1). The time course of forward posture holding and tipping tasks showed a significant decrease in tremor within 60 minutes after stimulation ( Figure 7 C, Figure 7 D, Table 1). No adverse events were reported.

[0134] For the finger-to-nose action task ( Figure 8 A; R 2 = 0.40, p < 10 -4 ), spiral plotting ( Figure 8 B; R 2 = 0.57, p < 10 -4 ), forward posture holding ( Figure 8 C; R 2 = 0.58, p < 10 -4 ), a significant correlation was found between the average peak tremor power (power 峰值平均值 ) and clinical visual assessment, while the degree of tipping was less ([[]]Figure 8 D). Figure 8 A - Figure 8 Each point in D indicates the value of a single segment. Peak tremor power is the power of wrist acceleration between 4 and 12 Hz (average power of a window around the peak frequency ±1.5 Hz). The clinical ratings in the graph have been jittered to improve the visibility of each point.

[0135] Figure 9 A - Figure 9 D illustrates that the kinematics measured by the sensor can predict clinical ratings across tasks. For each sub - panel ( Figure 9 A - Figure 9 D), the left column shows the normalized confusion matrix for prediction from a model trained using 50% of the data in each task. For the segments in each true clinical rating group, the values indicate the percentage of segments with the predicted rating. Prediction accuracy, balanced for the size of the clinical rating groups, measured by mean absolute error (MAE): Figure 9 A, Action task = 0.62, Figure 9 B, Drawing task = 0.78, Figure 9 C, Posture - holding task = 0.42 and Figure 9 D, Pouring task = 0.54. The right column shows the importance of the model input features, sorted by the information gain of each feature.

[0136] Shown in Figure 9 A - Figure 9 D, the predicted clinical ratings, as shown diagonally and non - diagonally, have the highest values in the confusion matrix ( Figure 9 left column) reflecting their true clinical ratings. Across tasks, the prediction accuracy of the model ranges from ±0.42 - 0.81 clinical rating units. This indicates that the model can unexpectedly and favorably predict clinical ratings, on average within ±0.62 of the actual clinical rating units, where the prediction for the posture - holding task is the most accurate (±0.42 clinical units). The output of the model can also be expressed as a probability. In some cases, the output of the model can indicate the length of time that a therapy will be effective before another stimulatory therapy is needed. Thus, the model improves the user's therapy and treatment. The model can also improve the nerve stimulation device by determining the proper usage of the device.

[0137] The distribution of prediction accuracy was determined, and how it is affected by the training proportion. The prediction accuracy of the model is not strongly affected by the amount of training data. By repeatedly building the model using different proportions of data for training and testing (for 80% training data, MAE: 0.48 - 0.64), it was found that for each task the prediction accuracy increases with a larger proportion of training data (R - 0.29 to - 0.1, p < 0.001, and the explained variance is 0.01 to 0.08). However, the improvement is small, indicating that even with a small training set, the model is still reliable and not overfitting.

[0138] It was found that the tremor reduction during and after median and radial nerve stimulation at the wrist lasted as long as the measurement, lasting about or at least about 60 minutes after about or at least about 40 minutes of stimulation. Across different FTM - CRS tremor tasks (including movement, drawing, posture holding, and dumping tasks), significant and sustained tremor reduction was evident. These results show unexpectedly excellent and lasting effects after a single stimulation segment.

[0139] Motion sensing technology may be advantageous in monitoring tremor status. Sensor kinematic features mapped to clinical visual assessment can reduce the subjectivity and variance in assessing tremor severity and provide insights into the tremor status. Unexpectedly and advantageously, it was found that the tremor severity measured with an accelerometer on the wrist is correlated with visual assessment. In some cases, angular velocity may also be correlated with visual assessment. Although other tasks such as those disclosed herein can alternatively or combinatorially be utilized, certain tasks such as posture holding are particularly advantageous in some cases for using mobile sensors to capture information regarding tremor severity.

[0140] The systems and methods herein are capable of predicting clinical assessment, on average within ±0.48 - 0.64 of the clinical assessment (which is a meaningful resolution for home measurements). The observed prediction error is less than the resolution of the clinical assessment scale (1 unit), and it is as accurate as or better than a human assessor. The prediction accuracy and the importance of kinematic features for prediction vary with the amount of training data, but the effect is weak (explained variance is 0.01 to 0.08), indicating that the prediction accuracy is not sensitive to small training datasets. The features used for predicting clinical assessment in these models can also be used to predict treatment response over time and at home. For example, Figure 9 The feature is shown therein.

[0141] In some embodiments, neuromodulation therapies can be combined with real-time wrist kinematic measurements to treat and predict tremor reduction in a clinic or other setting. As an example, median and radial nerve stimulation at the wrist has a lasting effect on essential tremor, lasting at least about 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 15, 18, 24 hours or longer after stimulation, or a range including any two of the above values. Non-invasive neuromodulation can be combined with sensor data recorded immediately before, during, and after stimulation to understand treatment response (e.g., in a home environment).

[0142] Figure 10 It is a graph showing the correlation between movement data (e.g., accelerometer) and a clinical scale (the Essential Tremor Rating Assessment Scale for Tremor (TETRAS)). When the patient performed a tremor-inducing task, accelerometer data of tremor movements were collected in three 12-second recordings, while a trained doctor performed the TETRAS assessment simultaneously. The tremor severity from the accelerometer was calculated as tremor power and plotted on a logarithmic scale. To quantify the tremor power, the average power was calculated in the frequency range of 4 - 12 Hz in the frequency band corresponding to the strongest tremor oscillation in the accelerometer signal (peak power frequency + / - 1.5 Hz).

[0143] Figure 11 Shows that collecting data over a longer time period can improve the prediction of patient response. In some embodiments, patient response can be predicted after a therapy session. However, in some embodiments, patient response is not predicted after a therapy session, and a therapy session of about or at least about 1 week, 2 weeks, 3 weeks, 1 month, 2 months, 3 months, or longer, and / or 2, 3, 4, 5, 6, 7, 8, 9, 10, 15, 20, 25, 30, 35, 40, 45, 50, or a longer therapy session of a specified length (e.g., in some embodiments, 30 minutes, 40 minutes, 60 minutes, or more or less) may be required. Figure 11 Plotted the correlation of patient response averaged over 1 day / 1 week / 1 month during a 3-month study and patient response randomly selected from 1 day / 1 week / 1 month results, showing that the correlation with patient response improves over time from 1 day to 1 week to 1 month.

[0144] Figure 12 Further shows how patient response on the first day / first week / first month of therapy predicts response for all sessions based on the average results during a 3-month study, and shows that the correlation with patient response improves over time from 1 day to 1 week to 1 month.

[0145] Figure 13It is illustrated that the efficacy of neuromodulation therapy is high for patients during drug (e.g., tremor drug) therapy and for patients not on drug therapy, and especially for patients not on drug therapy.

[0146] Figure 14 It is a graph showing that the therapy outcome measured for a patient is related to the kinematic improvement measured by a sensor, showing that the average PSI (patient self-report of improvement) after each segment is related to the kinematic improvement, as measured by the median improvement rate (log10(pre / post)==fold change). In some embodiments, the correlation between kinematic improvement and other user satisfaction / assessment metrics (e.g., NPS, QUEST) can also be achieved using the systems and methods disclosed herein.

[0147] Figure 15 A- Figure 15 D illustrates an example of endpoints for sensor metrics according to some embodiments. Figure 15 A illustrates representative time series and combined spectral data recorded before (top) and after (bottom) a single stimulus. As Figure 15 B illustrates the visual assessment correlation with simultaneous sensor recordings during postural hold. Figure 15 C illustrates the individual average percentage improvement in tremor severity of a subject at home over three months and the average improvement at the population level. Figure 15 D illustrates the average change in tremor severity per segment for all subjects, shown on a logarithmic scale.

[0148] Terms

[0149] When a feature or element is referred to herein as being "on" another feature or element, it can be directly on the other feature and / or element, or there can also be intervening features and / or elements. Conversely, when a feature or element is referred to as being "directly" on another feature or element, there are no intervening features or elements. It will also be understood that when a feature or element is referred to as being "connected", "attached" or "coupled" to another feature or element, it can be directly connected, attached or coupled to the other feature or element, or there can be intervening features or elements. Conversely, when a feature or element is referred to as being "directly connected", "directly attached" or "directly coupled" to another feature or element, there are no intervening features or elements. Although described and illustrated with respect to one embodiment, the features or elements described or illustrated can be applied to other embodiments. Those skilled in the art will also understand that a reference to a structure or feature being "adjacent" to another feature can have portions that overlap or lie beneath the adjacent feature.

[0150] The terms used herein are for the purpose of describing particular embodiments only and are not intended to limit the present invention. For example, as used herein, the singular forms "a", "an" and "the" are also intended to include the plural forms unless the context clearly indicates otherwise. It will also be understood that when used in this specification, the terms "comprises" and "comprising" specify the presence of the stated features, steps, operations, elements and / or components, but do not preclude the presence or addition of one or more other features, steps, operations, elements, components and / or groups thereof. As used herein, the term "and / or" includes any and all combinations of one or more of the associated listed items and may be abbreviated as " / ".

[0151] For ease of description, spatially relative terms, such as "below", "beneath", "lower", "above", "upper", etc., may be used herein to describe the relationship of one element or feature to another or other elements or features as illustrated in the figures. It will be understood that the spatially relative terms are intended to encompass different orientations of the device in use or operation in addition to the orientation depicted in the figures. For example, if the device in the figures is inverted, an element described as "below" or "beneath" another element or feature will be oriented "above" the other element or feature. Thus, the exemplary term "below" can encompass both an orientation of "above" and "below". The device may be otherwise oriented (rotated 90 degrees or at other orientations), and the spatially relative descriptors used herein are to be interpreted accordingly. Similarly, the terms "upwardly", "downwardly", "vertical", "horizontal", etc. are used herein for explanatory purposes only unless specifically stated otherwise.

[0152] Although the terms "first" and "second" may be used herein to describe various features / elements (including steps), these features / elements should not be limited by these terms unless the context indicates otherwise. These terms may be used to distinguish one feature / element from another. Thus, without departing from the teachings of the present invention, a first feature / element discussed below may be termed a second feature / element, and similarly, a second feature / element discussed below may be termed a first feature / element.

[0153] Throughout the specification and the appended claims, unless the context requires otherwise, the word "comprises" and variations such as "comprises" and "comprising" are meant to include various components that may be applied jointly in methods and articles (e.g., compositions and devices including apparatus and methods). For example, the term "comprises" will be understood to imply the inclusion of any stated element or step but not the exclusion of any other element or step.

[0154] As used in this specification and the claims, including as used in the examples, and unless expressly specified otherwise, all numbers may be read as if prefaced by the word "about" or "approximately", even if the term does not expressly appear. When describing magnitudes and / or positions, the phrase "about" or "approximately" may be used to indicate that the described value and / or position is within a reasonable expectation range of the value and / or position. For example, a numerical value may have a value (or range of values) of + / -0.1% of the stated value, a value (or range of values) of + / -1% of the stated value, a value (or range of values) of + / -2% of the stated value, a value (or range of values) of + / -5% of the stated value, a value (or range of values) of + / -10% of the stated value, etc. Any numerical value given herein should also be understood to include about or approximately that value, unless the context indicates otherwise. For example, if the value "10" is disclosed, then "about 10" is also disclosed. Any numerical range recited herein is intended to include all sub-ranges subsumed therein. It should also be understood that when a value of "less than or equal to" a value is disclosed, values of "greater than or equal to that value" and possible ranges between the values are also disclosed, as appropriately understood by those skilled in the art. For example, if the value "X" is disclosed, then "less than or equal to X" and "greater than or equal to X" are also disclosed (e.g., where X is a numerical value). It should also be understood that throughout the application, data is provided in many different formats, and that this data represents endpoints and starting points, as well as ranges for any combination of data points. For example, if a particular data point "10" and a particular data point "15" are disclosed, then it should be understood that values greater than, greater than or equal to, less than, less than or equal to, and equal to 10 and 15 are considered disclosed, as well as ranges between 10 and 15. It should also be understood that each unit between two particular units is also disclosed. For example, if 10 and 15 are disclosed, then 11, 12, 13, and 14 are also disclosed.

[0155] Although various illustrative embodiments have been described above, many variations of the various embodiments may be made without departing from the scope of the invention as described by the claims. For example, in alternative embodiments, the order in which the various described method steps are performed may often be changed, and in other alternative embodiments, one or more method steps may be skipped altogether. Optional features of various apparatus and system embodiments may be included in some embodiments and not included in other embodiments. Accordingly, the foregoing description has been provided primarily for purposes of illustration and should not be construed as limiting the scope of the invention as set forth in the claims.

[0156] The examples and illustrations included in this document show, by way of illustration and not limitation, specific embodiments in which the subject matter may be practiced. As mentioned, other embodiments may be utilized and other embodiments may be derived therefrom, such that structural and logical substitutions and changes may be made without departing from the scope of the disclosure. If more than one invention or inventive concept is in fact disclosed, for convenience only and not by way of a voluntary limitation of the scope of this application to any single invention or inventive concept, such embodiments of the inventive subject matter may be referred to herein individually or collectively by the term "invention". Thus, while specific embodiments have been shown and described herein, any arrangement calculated to achieve the same purpose may be substituted for the specific embodiments shown. The disclosure is intended to cover any and all modifications or variations of various embodiments. By reading the above description, combinations of the above embodiments and other embodiments not specifically described herein will be apparent to those skilled in the art. The methods disclosed herein include certain actions taken by a practitioner; however, whether explicitly or implicitly, they may also include any third-party instructions for those actions. For example, an action such as "stimulating the afferent peripheral nerve via skin irritation" includes "directing the stimulation of the afferent peripheral nerve".

Claims

1. A wearable nerve stimulation device for transcutaneously stimulating one or more peripheral nerves of a user, the device comprising: One or more electrodes configured to generate an electrical stimulation signal; One or more sensors configured to detect movement signals, wherein the one or more sensors are operably connected to the wearable nerve stimulation device; And One or more hardware processors configured to: Receive a raw signal in the time domain from the one or more sensors; Separate the raw signal into a plurality of frames; For each of the plurality of frames: Transform the raw signal into the frequency domain; Calculate a first energy in a first frequency band of the transformed signal for each frame; Calculate a second energy in a second frequency band of the transformed signal for each frame, wherein the second frequency band includes a first frequency corresponding to tremor; Determine movement artifacts in each frame based on a comparison of the first energy and the second energy; Combine frames based on the determination of movement artifacts for each of the frames; Extract features from the combined frames in the time domain or the frequency domain; Determine rules based on the extracted features; And Determine a nerve stimulation therapy outcome based on an application of the determined rules to operational data.

2. The wearable nerve stimulation device according to claim 1, wherein the sensor is operably attached to the wearable nerve stimulation device.

3. The wearable nerve stimulation device according to claim 1, wherein the raw signal is related to the tremor activity of the user.

4. The wearable nerve stimulation device according to claim 1, further comprising one or more end effectors configured to generate stimulation signals other than electrical stimulation signals.

5. The wearable nerve stimulation device according to claim 4, wherein the stimulation signal other than the electrical stimulation signal is a vibration stimulation signal.

6. The wearable nerve stimulation device according to claim 1, wherein the sensor comprises one or more of a gyroscope, an accelerometer, and a magnetometer.

7. The wearable nerve stimulation device according to claim 1, wherein the first frequency band is between about 0 Hz and about 2.5 Hz.

8. The wearable nerve stimulation device according to claim 1, wherein the second frequency band is between about 4 Hz and about 12 Hz.

9. The wearable nerve stimulation device according to claim 1, wherein the second frequency band is between about 3 Hz and about 8 Hz.

10. The wearable nerve stimulation device according to any one of claims 1 to 9, wherein the feature comprises at least one or more of the following: amplitude, bandwidth, area under the curve, energy in a frequency bin, peak frequency, or ratio between frequency bands.

11. The wearable nerve stimulation device according to any one of claims 1 to 9, wherein the feature comprises at least one or more of the kinematic features, and the kinematic features comprise the regularity, amplitude, and shape of the signal.

12. The wearable nerve stimulation device according to any one of claims 1 to 9, wherein the feature comprises at least one or more of the following: amplitude or power spectral density (PSD) at the peak tremor frequency, total amplitude or PSD in a frequency band approximately 2.75 Hz wide around the peak tremor frequency, total amplitude or PSD in a frequency band between approximately 4 Hz and approximately 12 Hz, or total amplitude or PSD in a frequency band around the peak tremor frequency selected only from the pre - stimulation spectrum.

13. The wearable nerve stimulation device according to any one of claims 1 to 9, wherein the feature comprises frequency - domain features.

14. The wearable nerve stimulation device according to any one of claims 1 to 9, wherein the feature comprises at least one of approximate entropy, displacement, curve fitting, functional PCA, filtering, mean, median, or time - domain extent.

15. The wearable nerve stimulation device according to any one of claims 1 to 9, wherein the feature comprises time - domain features.

16. A wearable device for transcutaneously modulating one or more peripheral nerves of a user, the device comprising: One or more end effectors configured to generate a stimulation signal; One or more sensors configured to detect movement signals, wherein the one or more sensors are operably connected to the wearable device; And One or more hardware processors configured to: Receive a raw signal in the time domain from the one or more sensors; Separate the raw signal into a plurality of frames; For each of the plurality of frames: Transform the raw signal into the frequency domain; Calculate a first energy in a first frequency band of the transformed signal for each frame; Calculate a second energy in a second frequency band of the transformed signal for each frame, wherein the second frequency band includes a first frequency corresponding to tremor; Determine movement artifacts in each frame based on a comparison of the first energy and the second energy; Combine frames based on the determination of movement artifacts for each of the frames; Extract features from the combined frames in the time domain or the frequency domain; Determine rules based on the extracted features; And Determine one or more of a clinical score and a nerve stimulation therapy outcome based on an application of the determined rules to operational data.

17. The wearable device according to claim 16, wherein the one or more hardware processors are configured to determine a first calibration frequency for a first stimulation therapy during a first activity and a second calibration frequency different from the second stimulation therapy during a second different activity.

18. The wearable device according to claim 17, wherein the first calibration frequency is within about 3 Hz of the second calibration frequency.

19. The wearable device according to claim 16, wherein the first frequency band is between about 0 Hz and about 2.5 Hz.

20. The wearable device according to claim 16, wherein the second frequency band is between about 4 Hz and about 12 Hz.

21. The wearable device according to claim 16, wherein the second frequency band is between about 3 Hz and about 8 Hz.

22. The wearable device according to claim 16, wherein the feature includes at least one or more of the following: amplitude, bandwidth, area under the curve, energy in a frequency bin, peak frequency, or ratio between frequency bands.

23. The wearable device according to claim 16, wherein the feature includes at least one or more of the kinematic features, and the kinematic features include the regularity, amplitude, and shape of the signal.

24. The wearable device according to claim 16, wherein the feature includes at least one or more of the following: amplitude or power spectral density (PSD) at the peak tremor frequency, total amplitude or PSD in a frequency band about 2.75 Hz wide around the peak tremor frequency, total amplitude or PSD in a frequency band between about 4 Hz and about 12 Hz, or total amplitude or PSD in a frequency band around the peak tremor frequency selected only from the pre-stimulation spectrum.

25. The wearable device according to claim 16, wherein the feature includes a frequency domain feature.

26. The wearable device according to claim 16, wherein the feature includes at least one of approximate entropy, displacement, curve fitting, functional PCA, filtering, mean, median, or time domain range.

27. The wearable device according to claim 16, wherein the feature includes a time domain feature.

28. A wearable nerve stimulation device, comprising: One or more electrodes configured to generate an electrical stimulation signal; One or more sensors configured to detect movement signals, wherein the one or more sensors are operably connected to the wearable nerve stimulation device; And One or more hardware processors configured to: Determine a first feature, the first feature including a first frequency in a 4 - 12 Hz frequency band having the highest power; Determine a second feature, the second feature including a first power at a peak of a frequency having the highest power in the 4 - 12 Hz frequency band; Determine a third feature, the third feature including an average power in a ±1.5 Hz window centered at the frequency having the highest power in the 4 - 12 Hz band; Determine a fourth feature, the fourth feature including a total power in the ±1.5 Hz window centered at the frequency having the highest power in the 4 - 12 Hz band; Determine a fifth feature, the fifth feature including a total power in the 4 - 12 Hz band; And Determine a sixth feature, the sixth feature including an entropy of a power spectral density in the 4 - 12 Hz band; Determine a seventh feature including a Q factor, the Q factor being a peak frequency divided by a frequency range where the power spectral density is higher than 50% of the highest power in the 4 - 12 Hz band; Determine an eighth feature including a temporal regularity of time - series data; and Predict a therapeutic efficacy of the wearable nerve stimulation device based on an application of respective weights corresponding to each of the first feature, the second feature, the third feature, the fourth feature, the fifth feature, the sixth feature, the seventh feature, and the eighth feature.

29. The wearable nerve stimulation device according to claim 28, wherein the respective weights are calculated based on training using a machine learning model.

30. The wearable nerve stimulation device according to claim 28 or 29, wherein the therapeutic efficacy includes clinical assessment.

31. The wearable nerve stimulation device according to claim 28 or 29, wherein the therapeutic efficacy includes probability.

32. The wearable nerve stimulation device according to claim 28 or 29, wherein the therapeutic efficacy includes the time before the next treatment is required.

33. A nerve modulation device for modulating one or more nerves of a user, the device comprising: One or more electrodes configured to generate a neuromodulation signal; One or more sensors configured to detect motion signals; And One or more hardware processors configured to: Receive a raw signal in the time domain from the one or more sensors; Separate the raw signal into a plurality of frames; For each of the plurality of frames: Transform the raw signal into the frequency domain; Calculate a first parameter in a first band of the transformed signal for each frame; Calculate a second parameter in a second band of the transformed signal for each frame, where the second band includes a first frequency corresponding to a physiological characteristic of a user; Determine motion artifacts in each of the frames based on a comparison of the first parameter and the second parameter; Combine the frames based on the determination of motion artifacts for each of the frames; Extract features from the combined frames in the time domain or the frequency domain; Determine rules based on the extracted features; And Determine a neuromodulation therapy result based on an application of the determined rules on operational data.

34. The device according to claim 33, wherein the first frequency band is between about 0 Hz and about 2.5 Hz.

35. The device according to claim 33, wherein the second frequency band is between about 4 Hz and about 12 Hz.

36. The apparatus according to claim 33, wherein the second band is between about 3 Hz and about 8 Hz.

37. The apparatus according to claim 33, wherein the feature comprises at least one or more of the following: amplitude, bandwidth, area under the curve, energy in a frequency bin, peak frequency, or ratio between bands.

38. The apparatus according to claim 33, wherein the feature comprises at least one or more of the kinematic features, wherein the kinematic features comprise the regularity, amplitude, and shape of the signal.

39. The apparatus according to claim 33, wherein the feature comprises at least one or more of the following: the amplitude or power spectral density (PSD) at the peak tremor frequency, the total amplitude or PSD in a band about 2.75 Hz wide around the peak tremor frequency, the total amplitude or PSD in a band between about 4 Hz and about 12 Hz, or the total amplitude or PSD in a band around the peak tremor frequency selected only from the pre-stimulus spectrum.

40. The apparatus according to claim 33, wherein the feature comprises a frequency domain feature.

41. The apparatus according to claim 33, wherein the feature comprises at least one of approximate entropy, displacement, curve fitting, functional PCA, filtering, mean, median, or time domain extent.

Images